ANOTHER LOOK AT NOMINAL COMPOUNDS 
Pierre Isabelle 
D~partement de linguistique 
Universit~ de Montreal 
C.P. 6128, Succ. A, Montreal, Qua., Canada H)C )37 
ABSTRACT 
We present a progress report on our research 
on nominal compounds (NC's). Recent approaches to 
this probiem in linguistics and natural ianguage 
processing (NLP) are reviewed and criticized. We 
argue that the notion of "roie nominal", which is 
at the interface of linguistic and extraiinguis- 
tic knowledge, is crucial for characterizing NC'e 
as weII as other Iinguistic phenomena. We examine 
a number of constraints on the semantic interpre- 
tation ruies for NC's. Proposals are made that 
shouid improve the capability of NLP systems to 
deaI with NC's. 
I INTRODUCTION 
A. Problem Statement 
As a first approximation, we define a 
"nominal compound (NC) as s string of two or more 
nouns having the same distribution as a singie 
noun, as in example (I): 
(1) aircraft bomb bay door actuating cylinders 
We will see below that provisions have to be made 
in some cases for intervening adjectives. 
NC's can exhibit various degrees of lexica- 
lization, but we will focus our attention on 
productive rules for forming novel compounds. As 
to their surface syntax, NC's can assume any 
structure generated by the rule N --> N N; accor- 
dingly, their structural ambiguity grows exponen- 
tially with their length (following the "Catalan 
sequence"). How, then, do we determine that the 
normal interpretation for (i) imposes the bracke- 
ting shown in (2), rather than any of the other 
41 syntactically possible bracketings? 
(2) ((aircraft ((bomb bay) door)) 
(actuating cylinder)) 
B. Goal__ssof th___ee study 
We believe that the analysis of NC's repre- 
sents an important and largely unsolved problem. 
From a theoretical point of view this problem 
raises the question of how to deal with noun 
semantics. And since noun meaning appears to be 
closely connected with knowledge of the world, 
one is led to explore the modes of interaction 
between linguistic and conceptual knowledge. 
From an NLP perspective, NC's have turned 
out to be an important stumbling block for sys- 
tems that attempt to deaI with real-iife text, 
especiaIly in technical domains, for purposes 
such as machine translation (IsabeIle, to 
appear), information retrieval, etc. 
Our ultimate goal is to develop an NLP sys- 
tem capable of analyzing large classes of NC's in 
the sublanguage (Kittredge and Lehrberger, i982) 
of aircraft maintenance manuals. We do not aim at 
solving all cases, since we believe the problem 
to be exceedingly difficult. At the present stage 
of our inquiry, we concentrate on the design of s 
suitable theoretical framework. 
In section II, we present a brief review of 
previous work in linguistics and NLP. In sections 
III and IV, we examine two aspects of the seman- 
tics of nouns that are crucially relevant to the 
analysis of NC's: predicative nouns and role 
nominals. Finally, in section V, we explore pos- 
sible constraints on the semantic interpretation 
of NC's. 
II BACKGROUND 
A. Approaches in Linguistics 
The early study of Lees (1963) classified 
NC's on purely grammatical criteria, and it 
failed to provide constraints that could explain 
how NC's are semantically interpreted. 
In s number of more recent studies, such as 
Levi (1978)9 there has been an attempt to view 
NC's as governed by tight semantic constraints. 
Thus, according to Levi, any novel NC realizes a 
pattern where either: a) the head noun is a 
deverbal naminalizstion and its modifier is 
interpreted as an argument of the related verb; 
or b) the two nouns are related by one of exactly 
nine deletable predicates ("is the cause of", "is 
for", etc.). 
This reductionist attempt has been criti- 
cized, most notably by Downing (1977), on the 
509 
grounds that the interpretation of NC's crucially 
involves pragmatic knowledge, and that numerous 
cases of NC's (such as thalidomide parents) will 
resist any analysis in terms of a closed set of 
relations. These criticisms have led theorists 
like Selkirk (1982) to adopt the position that 
only "verbal compounds" (those constructed on a 
pattern "argument + nominalization") are amenable 
to linguistic characterization; all other NC's 
would have to be explained in separate extralin- 
guistic theories. 
C. Approaches in NLP 
Several systems have been developed in an 
NLP framework to deal with the problem of inter- 
preting NC's; two recent exempIes are reported in 
Finin (1980) and McDonaId (i982). 
In both systems, the individual nouns of an 
NC are first mapped onto conceptual representa- 
tions where concepts are characterized by a set 
of "roles" or "slots", and are arranged in an 
abstraction hierarchy. Interpreting a compound 
then amounts to forming a derived concept on the 
basis of the constituent concepts; in most cases, 
this is done by interpreting one of the concepts 
as a slot filler for the other. For example, the 
interpretation of steel adapter would involve 
inserting the concept associated with steel into 
a RAW-MATERIAL slot within the representation of 
adapter. But the authors do not examine in any 
detail the question of eventual constraints on 
this interpretation process, such as the effect 
of word order. 
Another crucial issue which has not been 
explored in sufficient detail, is the nature of, 
and the justification for the particular set of 
slots which is assigned to a given concept. Finin 
is somewhat more explicit on this question. Some 
nouns have slots which represent standard case 
roles. Role nominals, on the other hand, are 
nouns which refer to a particular case role of 
another concept; for example, food refers to the 
object role of eat, and this fact provides the 
key to the interpretation of cat food. This 
notion will be examined in detail below. But 
Finin also resorts to other types of slots (e.g. 
"raw-material") which are not discussed. 
III PREDICATIVE NOUNS 
A. Root Nouns with Arguments 
The fact that several classes of non-derived 
nouns strictly subcategorize phrases which are 
semantically interpreted as arguments has 
received very little attention in the literature. 
This phenomenon would deserve an extensive study, 
and we can only give here some relevant examples, 
together with an indication of the semantic cate- 
gories between which the relation expressed by 
the noun effects a mapping: 
(3) a. measure nouns 
map: objects onto quantities 
examples: speed (of), temperature (of), 
volume (of), size (of) 
b. "area" nouns 
map: objects onto subparts 
examples: top (of), side (of), bottom 
(of), center (of), core (of) 
c. collective nouns 
map: individuals onto sets 
examples: group (of), set (of) 
d. representational nouns 
map: objects onto representations of ob- 
jects 
examples: picture (of), diagram (of), 
sense (of) 
e. other examples 
location (of), goal (of), 
brother (of), king (of) 
Most if not all of these argument-taking nouns 
can have their argument satisfied by a modifier 
noun in an NC: 
(a) a. oil temperature 
b. box top 
c. tank group 
d. circuit diagram 
e. component location 
A treatment in terms of predicate/argument pat- 
terns for this type of NC seems far superior to 
Levi's (1978) use of the semantically empty 
"deletable predicate" HAVE ("the component HAS a 
location"). Although these NC's are excluded from 
Selkirk's (1982) class of verbal compounds, they 
are amenable to the same type of semantic des- 
cription. 
B. Action and State Nominalizations 
Most studies on NC's have recognized the 
fact that deverbal nominalizations exhibit a 
semantic behavior closely related to that of the 
verb, and subcategorize elements that can occur 
as modifier nouns in NC's. As mentioned above, 
these are in fact the only cases that Selkirk 
(1982) deems characterizable at the level of 
linguistic competence. 
8ut there seems to be no reason why deadjec- 
tival nominalizations should be handled in a 
different way. In examples (5) and (6), an action 
and a state are nominalized, with the argument 
occurring either in a prepositional phrase or as 
a modifier in an NC: 
(5) a. Someone removes the pump. 
b. removal of the pump 
c. pump removal 
(6) a. Uranium is scarce. 
b. scarcity of uranium 
c. uranium scarcity 
510 
We are not claiming that there is exact 
synonymy between the (b) and (c) examples, but 
only that they exhibit the same predicate/ 
argument pattern. Action nominals can take 
various types of arguments in NC's, and sometimes 
several of them simultaneously: 
(7) a. pump failure (subject) 
b. Montreal flight (source, goal) 
(8) a. poppet chatter tendency 
IV ROLE NOMINALS 
A. Nominalizations 
Deverbal nominalizations can refer not only 
to the action expressed by the verb, but also to 
the agent (driver), instrument (lubricant), pa- 
tient (employ-~-~and result (assembly) of this 
action. Except maybe for results, the term "role 
nominal" seems appropriate, since the nominaliza- 
tion refers to the filler of one of the roles of 
the verb. Although these nouns are not, strictly 
speaking, predicative, they generally permit to 
form NC's in which the other is interpreted as an 
argument of the underlying verb: 
(9) truck driver 
= one who drives trucks 
(I0) engine lubricant 
= something with which engines are 
cated 
ill) 18H employee 
= one who is employed by IBH 
lubri- 
(12) pump assembly 
= the result of assembling a pump 
With agent and instrument nominals (9, i0), there 
is a strong tendency to assign an "habitual" 
aspect to the underlying verb and generic refe- 
rence to its object; this kind of interpretation 
is awkward when the argument appears in a PP: 
(I)) ?a driver of trucks 
B. Root Nouns 
The term "role nominal" is due to Finin 
(1980) who uses it to cover not only nominaliza- 
tions of the type described above, but also any 
noun which can be semantically interpreted as 
referring to a role of a given verb, whether or 
not this verb happens to be morphologically 
related. This claims amounts to saying that, 
semantically, we have relations such as the 
following: 
(14) a. pilot:fly :: driver:drive 
b. gun:shoot :: lubricant:lubricate 
c. food:eat :: employee:employ 
A related claim underlies Zholkhovskij & 
Hel'cuk's (1971) use of certain "lexical func- 
tions": 
(15) a. Sl(bUy) = buyer 
b. S3(buy) = seller 
c. Sloc(battle) = battlefield 
d. Sinst(See) = eyes 
where Si, $1o c and Sinst are defined as functions 
t~t yield the typical name of, respectively, the 
i- arrant, the location a-nd the instrument. 
Fillmore (1971) also makes a comparable 
proposal, when he suggests that the lexical entry 
of knife should include (16) as a component: 
(16) use: I of <V 0 1 A> 
where V=cut 
C. On the Definition of Role Nominals 
Mow exactly should we understand the notion 
of role nominals? Finin's statement that they 
refer to an underlying case role of a verb is not 
accurate: they refer to role fillers, not to the 
roles themselves. Assuming that they denote a set 
of role fillers, we can ask if this set is: 
a. the set of all possible fillers for the 
role; or 
b. a set of typical fillers for that role; 
or 
c. any set of possible fillers for the role. 
Possibility (a) seems to describe correctly 
numerous cases of deverbal nouns. For example, it 
seems clear that anyone who is employed is an 
employee. However, with agents and instruments, 
there is a tendency to reserve the role nominal 
for habitual fillers: one hesitates to apply the 
term writer to a person who only wrote a letter. 
Horeover, with this definition, knife would not 
be a role nominal for cut, since it is perfectly 
possible to cut bread with a sword. The notion 
would then loose much of its power, since we need 
it to explain why bread knife is interpreted as 
"a knife used as an instrument for cutting 
bread". 
On the other hand, definition (c) seems too 
weak: even if a sword can be used to cut bread, 
bread sword is odd in a normal context (contex- 
tual factors will be discussed below). Thus, 
definition (b) seems the most appropriate. 
D. 3uatifyin9 Role Assignments 
For those role nominals where there is no 
morphological evidence of relatedness with the 
underlying verb, one is forced to rely mostly on 
intuitions. However, the risk of arbitrariness 
can be reduced by looking for further evidence 
from other linguistic phenomena. 
511 
i. Evaluative Adjectives 
When Fillmore (1971) introduced his notion 
of role nominal, he was attempting to characte- 
rize the behavior of evaluative adjectives, not 
the behavior of NC's. He noted that a ~ood ~, 
where X is a role nominal, means: 
a. if X is an agent: one who performs the 
associated activity skilfully 
(s good driver, a good pianist); 
b. if X is an instrument: a thing which 
permits the associated activity to be 
performed easily 
(a good knife, a good broom); 
c° In other cases, it seems that the resul- 
tant meaning is less predictible 
(good food has certain properties concer- 
ning nutritiousness and taste; a good 
house is comfortable, built to last, 
etc.). 
As far as we can tell, the evaluation domain 
for agents and instruments is precisely the acti- 
vity which is relevant for the understanding of 
NC'B. Thus while good driver evaluates the dri- 
ving, car driver specifies its object; moreover, 
in ~ood car driver, car falls within the evalua- 
tion domain: car drivers and truck drivers are 
evaluated on different scales. Evaluative adjec- 
tives can thus be used as a further source of 
evidence in the description of role nominals. 
2. Denominal Verbs 
Another phenomenon which is relevant to the 
question of role nominals and NC's is the crea- 
tion of denominal verbs. Clark & Clark (1979) 
examine this very productive process, in which a 
verb formed by zero-affixation is understood "in 
such a way that the parent noun denotes one role 
in the situation, and the remaining surface argu- 
ments of the denominal verb denote other roles in 
the situation" (p. 787). For example, Max sub- 
wayed downtown means that Max went downtown on 
a subway. Intuitively, it appears obvious that 
the knowledge involved in this interpretation 
process is very closely related to whatever per- 
mits interpreting the NC downtown subway as "the 
subway that goes downtown". 
An important aspect of C & C's work is to 
show that the formation of denominal verbs is 
heavily dependent on contextuai knowIedge. Thus 
if you and me both know that Phii has iong had 
the crazy habit of sticking trombones into the 
nose of bypassers, I can inform you that PhiI has 
just tromboned a poiice officer. Notice that in 
the same context, 9opd trombone wouId presumably 
mean a trombone that is easy to stick into some- 
one's nose. 
In such cases, the interpretation is based 
on particular, situational knowledge. NC's can 
also be based on this type of knowledge. For 
example, if that same Phil uses different types 
of trombones for men and women, we might speak of 
his women trombones, to mean trombones of the 
type that Phil sticks into the nose of female 
bypaasers. 
But C & C claim that, more frequently, deno- 
minal verbs are based on ~eneric knowledge about 
concrete objects, knowIedge which is accessibIe 
to aii speakers in a Iinguistic community. This 
claim is to be linked with Downing's (i977) 
remark that aithough NC's are sometimes "deic- 
tic", they are most often based on generic or 
permanent reIationships. 
Obviously, the relevant knowledge (whether 
particular or generic) is at least as much about 
the world as about language. In fact, it is clear 
that both types condition each other. For exam- 
ple, objects that are used as vehicles will tend 
to be verbalized on the syntactic pattern of 
movement verbs; and in the absence of other evi- 
dence, one is likely to infer from its syntax 
that The fairy pumpkined the kids to Narnia allu- 
des not to an "satin 9 pumpkin" but to a "trans- 
portation pumpkin". 
In order to predict the range of meaning of 
large classes of denominal verbs, C & C propose 
to encode some generic knowledge in the lexicon, 
by means of "predominant features" such as: 
(17) a. x is the agent of Act (to pilot y) 
b. x is the instrument of Act (to pump y) 
c. x is the result of Act (to group y) 
d. x is the location of y (to can y) 
e. x is the locatum of y (to cover y) 
f. x is the time of Act (to weekend in y) 
It is quite apparent that these features are 
meant to capture the same type of facts as role 
nominsIs. It is easy to find NC's which paraIiel 
each ciass of verb singled out by them: 
(17') a. aircraft piiot 
b. oii pump 
c. tank group 
d. oil can 
e. pump cover 
f. Montreal weekend 
We believe that the semantic mechanisms that 
are at work in the interpretation of NC's, deno- 
minal verbs and evaluative adjectives are basi- 
cally similar. By using independent evidence from 
these phenomena, and an adequate generalization 
of the notion of role nominal, one can go s long 
way toward uncovering the relevant semantic 
mechanisms. 
The notion of role nominal, as we understand 
it, is at the interface between linguistic and 
extralinguistic knowledge. Nouns may have severaI 
different roies, depending on the contingencies 
of the entities that they denote; in those cases 
context usuaiiy makes one roie more saiient. In 
fact, context may even impart a noun with an 
unusuai roie, as we have seen. However, we 
beiieve that in ordinary texts, such as technieai 
512 
manuals, NC's that are analyzable in terms of 
role nominals are most often based on the usual, 
generic roles of the nouns. But this can only be 
shown through a large scale description of the 
relevant NC's. 
E. A Tentative Scheme 
Since the knowledge we have to encode is 
tied to world knowledge, there is a risk that it 
could become overwhelmingly compiex. However, we 
will minimize this risk, by limiting the scope of 
our inquiry to a suitably restricted subianguage. 
Technical manuals are interesting in this respect 
because they exhibit at the same time a tightIy 
constrained universe of interpretation, and an 
exceptional productivity in compounding. 
As to the semantic framework, we will not 
use case grammar, even if our discussion of role 
nominals was couched in the terms of this theory 
for expository purposes. As is well-known, case 
grammars raise a number of difficult problems. 
For example, the distinction between agent 
and instruments is problematic. At the morpholo- 
gical level, both can give rise to -er nominali- 
zations; at the syntactic level, both can appear 
in subject position, and in with NP environments 
(cf. co-agents); and at the semantic level, both 
notions are frequently undistinguishable, espe- 
cially in texts dealing with machines, such as 
technical manuals. Since action verbs can gene- 
rally occur with an agent, an instrument, or 
both, it does not seem necessary to include two 
slots for each verb in the dictionary: general 
rules should predict the relevant facts. 
Another problem is that if we want the 
notion of role nominal to be general enough, a 
role nominal should be able to refer to entities 
which are not case slots, at least in the usual 
sense of this term. Result nominals, for instan- 
ce, do not refer to a case slot as such. 
There is in fact no evidence that the inter- 
pretation rules for NC's crucially involve case 
roles rather than argument places, and our de- 
scriptions will be couched in terms of a predi- 
cate/argument notation. This notation is perfec- 
tly compatible with the use of an ontology that 
is richer than standard predicate logic. 
For example, if we agree with 3ackendoff's 
(198)) claim that "place" and "paths" constitute 
basic cognitive categories, we can define "typed" 
variables with the appropriate range. In our 
descriptions below, p will denote a path varia- 
ble, that is, a variable ranging over entities 
denoted by complex expressions constructed out of 
,! ,, path functions (into, from, toward, via, etc.) 
,! ,, and places . Similarly, we will use e as an 
"event" variable, as in Moore (1981) ~d Hobbs 
(1984). 
A role nominal will contain within its lexi- 
cal entry one or more statements of the form "x 
such that P(x)". We do not think that diacritic 
markers such as "typical function" are required; 
rather, notions such as typicality, should be a 
consequence of the semantic rules which interpret 
lexical entries. 
We give below a few examples of the type of 
lexical specification for role nominals with 
which we want to experiment. At this stage of our 
work this should be taken as nothing more than a 
first approximation. The material enclosed in 
brackets represents selectional restrictions. 
(18) pilot 
x such that FLY(x,y,p) 
<aircraft(y)> 
(19) adapter 
x such that ADAPT(x,y,z) 
(this entry is produced 
morphology) 
by derivational 
(20) tank 
x such that CONTAIN(x,y) 
<fluid(y)> 
(container, chamber, reservoir, bay, compar- 
tment, etc. are similar but selection res- 
trictions can differ) 
(21) hinge 
x such that ATTACH(x,y,z) 
(22) brace 
x such that SUPPORT(x,y) 
(23) witness 
x such that SEE(x,e) 
(24) line 
x such that CONDUCT(x,y,p) 
<fluid(y)> 
IV CONSTRAINING INTERPRETATION RULES 
The preceding sections have discussed two 
aspects of the lexical semantics of nouns that 
are relevant for the analysis of NC's: argument- 
taking nouns and role nominals. Assuming that the 
lexicon contains this type of information, we can 
now ask how it is used by semantic interpretation 
rules. 
More specifically, the picture that emerges 
is that lexical entries provide predicate/argu- 
ment patterns which contain variables; these 
variables have to be bound to the semantic mate- 
rial associated with some other noun. We must 
then ask what are the rules that govern this 
binding process. 
513 
A. Relative Order of the two Nouns 
I. Predicative Nouns 
Selkirk (1982) claims that the modifier noun 
can satisfy an argument of the headnoun but not 
vice-versa. Finin (1980) claims that argument 
satisfaction (or slot filling) is possible in 
both directions. Finally, McDonald (1982) takes 
the intermediate position that satisfaction of 
the head by a modifier is much more frequent. 
Now, consider the following pairs of exam- 
ples: 
(25) a. oil temperature 
b. ??temperature oil 
(26) a. uranium scarcity 
b. ??scarcity uranium 
(27) a. pump removal 
b. ??removal pump 
(28) a. student invention 
b. ??invention student 
It is clear that the (b) examples, if interpreta- 
ble st all, cannot easiIy realize the pattern 
found in the (a) examples. However, there is a 
very productive process which permits an action 
nominal to occur to the left of its argument; 
this pattern is most productive with inanimate 
headnouns: 
(29) a. repair man 
b. ?man repair 
(30) a. cooling device 
b. device cooling 
(3l) a. bleed valve 
b. ??valve bleed 
c. valve bleeding 
(32) a. jacking point 
b. ??point jacking 
In the first three (a) examples, the headnoun is 
interpreted as a subject argument (agent or ins- 
trument). In (28a), point is interpreted as a 
iocation (where one jacks something). In all of 
the (a) examples, the action nominai is inter- 
preted as denoting a permanent roie or function 
of the headnoun. 
When order is reversed, meaning and, even- 
tually, acceptability are affected. In the (b) 
examples, the action nominal is not interpreted 
as expressing a permanent function of the other 
noun. Thus, if one can manage subject interpreta- 
tions in (29b) and (31c) -- "a repair done by men 
(not robots)" and "a valve which bleeds some- 
thing" --, it will not follow that the men are 
repair men or that the valve is a bleed valve. 
The pattern "argument + predicative noun" 
has some pecularities of its own; for example, 
why is (32b) unacceptable no matter the role 
assignment that one makes? Or why is girl swim- 
min 9 unacceptable? In the latter case, it cannot 
be, as suggested by Selkirk (1982) that subject 
arguments are prohibited in general: pump failure 
is fine. It may be that subjects are ruled out 
for -ing nominals, unless they express a result 
(cf. consumer spending). 
But on the whole, the pattern where the 
predicate comes first is much more constrained, 
since it permits oniy action nominaIs, and pro- 
duces a semantic resuit very simiiar to =oie 
nominais. Thus, a cooIing device and cooier 
denote very simiIar entities; the same is true of 
jacking system and jack. Notice that in exampies 
such as: 
()3) air temperature monitoring system 
monitorin 9 system forms a constituent, even if 
air temperature is understood as the object of 
of the monitoring; this is confirmed by the fact 
that (34) receives a similar interpretation: 
(34) monitoring system for air temperature 
It seems that this modification pattern has the 
effect of creating a role nominal out of its two 
constituents. 
2. Role Nominals 
Here again, the order of the nouns strongly 
conditions the resulting interpretation: 
(35) a. truck driver 
b. ??driver truck 
(36) a. oil pump 
b. pump oil 
(37) a. pump case 
b. ?case pump 
(38) a. equipment bay 
b. bay equipment 
In the (a) examples, the modifier noun is inter- 
preted as the object of the underlying verb 
(drive, transfer, detect, hold); here too, a 
permanent connection is established. A truck 
driver is a person whose (social) function is to 
drive trucks, and that person is still a truck 
driver when he/she drives a scooter. If you use 
an oil pump to pump your tomato juice, it still 
qualifies to be called an oil pump. 
But when the nouns are inverted, there is a 
change in meaning and, possibly, in acceptabili- 
ty. It is hard to find a sensible interpretation 
for driver truck. The most natural interpretation 
for pump oil is "oil used for lubricating pumps": 
in this case, oil has become the relevant role 
nominal. There is also a possibility for an 
interpretation such as "oil in the pump" or "oil 
coming from the pump". We are not sure how that 
type of interpretation should be produced. But it 
seems clear that oil can only become an actor in 
514 
a permanent function of the pump when pump is the 
headnoun. 
It is easy to see that the other examples 
exhibit s similar behavior. We therefore conclude 
that role nominals tend to establish permanent, 
functional connections with their modifiers only. 
When they modify a headnoun, either this noun 
will absorb them itself (if it is a predicative 
or role nominal) or else a looser connection such 
as a locative one will be created. 
B. Multiple Modification 
i. Argument Ordering 
When a predicative noun or role nominal in 
head position is modified by more than one argu- 
ment, it seems that ordering (39) will apply, as 
illustrated in the examples below= 
()9) time < paths < subject < object 
loc 
(40) s. computer fuel testing 
b. ??fuel computer testing 
(41) a. Montreal jet flights 
b. ??jet Montreal flights 
(42) a. evening Montreal trains 
b. ??Montreal evening trains 
2. A Broader Perspective 
So far, we have concentrated on predicative 
nouns and role nominals, thereby excluding 
various types of compounds. When we look at the 
broader picture, different problems are raised. 
First, (in technical manuals) we find examples 
with intervening adjectives as in (43): 
(43) (Remove) front cockpit right shelf aft con- 
trol panel. 
While the formation of NC's such as those we 
have examined in previous sections is usually 
considered to be a lexical process, there is some 
evidence that, in examples like (43), syntactic 
processes are at work. Notice that each one of 
the three main groups in (43) is referential= 
"the (...) panel of the (...) shelf of the (...) 
cockpit", much as if there was an implicit geni- 
tive marker. But genuine compounds (that is, 
those of syntactic category N) are usually consi- 
dered to be anaphoric islands (Levi, 1978). One 
possible explanation is that (43) should in fact 
be analyzed as containing three NP's. 
A comparison of the following examples pro- 
vides some support for that view= 
(44) a. Remove front cockpit right shelf. 
b. Remove right cockpit shelf. 
In (44a), cockpit has definite reference and is 
interpreted as no more than a location for the 
shelf. However, in (44b), when the scope of righ t 
includes shelf, cockpit becomes nonreferential; 
cockpit shelf then denotes a type of shelf. These 
facts seem to indicate that this scoping of 
right, in virtue of syntactic constraints, forces 
one to take cockpit shelf as something of cate- 
gory N; and when we do so, shelf becomes a role 
nominal whose argument receives the type of "per- 
manent function" interpretation discussed above. 
Core NC'B, those which establish argument 
connections with a predicative noun or role 
nominal, appear to form a tightly bound unit: 
(45) a. ??oil high temperature 
b. ??truck good driver 
c. ??equipment left bay 
Some adjectives, the "nonpredicating" adjectives 
of Levi (1978) can appear within core NC's. But 
they are in fact nouns in disguise, and they 
receive an argument interpretation; in (46), it 
is understood that one repairs a structure. 
(46) aircraft structural repairs 
Thus, if we are willing to make a distinc- 
tion between syntactic compounding (similar to 
genitive phrases) and core NC's, it can be 
claimed that true compounds form anaphoric 
islands and cannot be separated by adjectives. 
Finally, there are certain types of nominal 
modifiers (for which we have as yet no analysis 
to offer) whose degree of cohesion with the 
headnoun is somehow intermediate between nouns 
interpreted as arguments and nouns in separate 
(pseudo-genitive) NP's. For example, nouns which 
stand in a "material of" relation to the head 
cannot precede adjectives, but must precede argu- 
ments= 
(43) a. large steel tank 
b. ??steel large tank 
(44) a. steel fuel tank 
b. ??fuel steel tank 
The constraints that we have seen, and no 
doubt others yet to be discovered, are certainly 
significant from a theoretical point of view; but 
they also have practical value for NLP systems 
which have to deal with NC's. 
VI CONCLUSIONS 
We have discussed two classes of nouns that 
have particular importance in NC formation= pre- 
dicative nouns and role nominals. We have shown 
that the latter class is also relevant to the 
description of evaluative adjectives and denomi- 
nal verbs. A tentative framework for the descrip- 
tion of role nominals has then been proposed. 
Finally, we have seen that rules which interpret 
515 
NC's obey e number of constraints. A distinction 
has been made between "pseudo-genitive" construc- 
tions and core NC's, the latter forming a tightly 
bound unit with internal ordering constraints. 
Much work remains to be done on the issues 
that we have discussed here; it is likely that 
experimentation with actual descriptions on a 
larger scale will lead to several refinements and 
revisions. Moreover, as has been pointed out in 
the last section, our description still ignores a 
number of types of compounding, and further dif- 
ficulties are to be expected. Nonetheless, we are 
confident that our work will, in the near future, 
result in a small NLP system capable of analyzing 
a broad range of NC's. 
Selkirk E., The Syntax of Words, MIT Press, 1982. 
Zholkovskij A., Mel'cuk I., SUr la synthbse sd- 
mantique, in T.A. Informations, 11:2, 1971. 
VII REFERENCES 
Aronoff M., Contextuals, in Language, 56:4, 744- 
758, 1980. 
Clark E., Clark H. When Nouns Surface as Verbs, 
in Language, 55:4,, 767-811, 1979. 
Downing P., On the Creation and Use of English 
Compound Nouns, in Lanquaqe, 5):4, 810-842, 
1977. 
Fillmore C., Types of Lexicsl Information, in D. 
Steinberg and L. 3akobovits (eds.), Semantics, 
Cambridge Univ. Press, 197I. 
Finin T., The Semantic Interpretation of Compound 
Nominals, Coordinated Science Lab., Univ. of 
IIIinois, 1980. 
Hobbs 3., Sublanguage and Knowledge , paper pre- 
sented at the NYU Conference on Subianguages, 
3anuary 1984. 
Isabelle P., Machine Translation at the TAUM 
Group, paper presented at the Lugano Tutorial 
on Machine TraneIation, April i984, to appear. 
3ackendoff R., Semantics and Cognition , MIT 
Press, 198). 
Kittredge R., Lehrberger 3., Sublanguage: Studies 
of Language in Restricted Semantic Domains, De 
Gruyter, 1982. 
Lees R., The Grammar of English Nominalizations, 
Mouton, i96). 
Levi 3., Th__ee Syntax and Semantics of CompIex 
Nominals, Academic, 1978. 
McDonald D.B., Understandin 9 Noun Compounds, 
Carnegie-Meiion University, i982. 
Moore R., Problems in Logical Form, Tech. note 
241, SRI, 1981. 
516 
